Zoomable Interactive Time Series VisualizationMitar Milutinović (mitar@tnode.com)

Problem∙ concurrent visualization of multiple time series datasets in the browser∙ interactive, zoomable, minimal data transferred to the browser, support

for huge datasets

Motivation∙ Internet of Things time series datasets are becoming more and more

common∙ sensors networks, citizen science measurements with mobile devices∙ huge datasets, but trend of moving everything into the cloud∙ comparison between data, but data is big, even one dataset can be

incomprehensible and visualization crowded∙ loss of information when zooming out

Related work∙ imMens: Real-time Visual Querying of Big Data∙ Nanocubes for Real-Time Exploration of Spatiotemporal Datasets∙ Horizon Graphs∙ Stack Zooming for Multi-Focus Interaction in Time-Series Data Visual-

ization

Initial viewInitial view. Time series navigator below for selecting. Two parallel timeseries on each chart, three charts in parallel.

Design principles∙ in the browser∙ interactive, mouse interface∙ scalable, RESTful API∙ minimization of “operation mode” switching∙ no need to understand the technical background of the underlying sys-

tem

Approach∙ three components

– server side time series storage (Python + MongoDB)– HTTP RESTful interface (Python + Django)– client visualization (JavaScript + Highstock)

∙ layered downsampled datapoints at various granularity levels– both value and time downsampling operators– multiple downsampling operators at once– user gets data from the granularity levels most suitable for the time-

span requested∙ derived time series from another time series∙ selecting, panning, zooming, highlighting∙ parallel inter-locked series on one chart, parallel inter-locked charts

Future work∙ client performance improvements∙ other types of values supported

– graphs– discrete events

∙ sharing of visualization – unique URL for each visualization configura-tion

∙ other types of visualizing time series data– stacked graphs– areas– background color regions

∙ analyzing time series properties (auto-correlation) and exposing them

Detail

Zoomed viewHighest detail zoom without range information.

Range viewRange around the mean, but no highlighting enabled.